Integrated metagenomics and networkanalysis of soil microbial community ofthe forest timberlineJunjun Ding1,2, Yuguang Zhang1, Ye Deng3,4, Jing Cong1,6, Hui Lu1, Xin Sun2, Caiyun Yang4, Tong Yuan4,Joy D. Van Nostrand4, Diqiang Li1, Jizhong Zhou2,4,5 & Yunfeng Yang2

1Institute of Forestry Ecology, Environment and Protection, and the Key Laboratory of Forest Ecology and Environment of State ForestryAdministration, the Chinese Academy of Forestry, Beijing 100091, China, 2State Key Joint Laboratory of Environment Simulationand Pollution Control, School of Environment, Tsinghua University, Beijing 100084, China, 3CAS Key Laboratory of EnvironmentalBiotechnology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing 100085, China, 4Institutefor Environmental Genomics and Department of Botany and Microbiology, University of Oklahoma, Norman, OK 73019, USA,5Earth Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA, 6School of Minerals Processing andBioengineering, Central South University, Changsha 410083, China.

The forest timberline responds quickly and markedly to climate changes, rendering it a ready indicator.Climate warming has caused an upshift of the timberline worldwide. However, the impact on belowgroundecosystem and biogeochemical cycles remain elusive. To understand soil microbial ecology of thetimberline, we analyzed microbial communities via 16s rRNA Illumina sequencing, a microarray-based toolnamed GeoChip 4.0 and a random matrix theory-based association network approach. We selected 24sampling sites at two vegetation belts forming the timberline of Shennongjia Mountain in Hubei Province ofChina, a region with extraordinarily rich biodiversity. We found that temperature, among all of measuredenvironmental parameters, showed the most significant and extensive linkages with microbial biomass,microbial diversity and composition at both taxonomic and functional gene levels, and microbialassociation network. Therefore, temperature was the best predictor for microbial community variations inthe timberline. Furthermore, abundances of nitrogen cycle and phosphorus cycle genes were concomitantwith NH4

1-N, NO32-N and total phosphorus, offering tangible clues to the underlying mechanisms of soil

biogeochemical cycles. As the first glimpse at both taxonomic and functional compositions of soil microbialcommunity of the timberline, our findings have major implications for predicting consequences of futuretimberline upshift.

The forest timberline, a distinctive boundary between the continuous closed forest canopy to the treelessalpine zone1,2, has solicited substantial attention because it responds quickly and markedly to climatechange3. Formation of the forest timberline is typically associated with mean air temperature between 5.5

and 7.5uC during the growth season4, which is a major limiting factor for tree growth at high altitudes2. As it hasbeen recognized that global warming induces upshift of the timberline towards higher altitudes worldwide5, thesensitivity of the timberline to temperature makes the timberline a ready indicator of climate change6.

Several studies of the forest timberline have explored the underlying mechanisms forming spatial distributionpatterns and ecological processes involved in biogeochemical cycles3,7. At the global scale, temperature is adeterminative factor controlling the timberline2,7. At the local scale, environmental parameters and anthro-pogenic perturbation of the landscape affect formation of the timberline8. However, a majority of these studiesonly analyze plant communities and their responses to climatic dynamics9–11. Only a few studies have analyzedmicrobial activities in regard to microbial biomass, carbon efflux and decomposition rates12–14, neglecting micro-bial community composition and structure. Due to vast complexity of soil microbial communities and in-adequacy of scientific tools that accurately profile microbial community structure and function15, it remainsunclear whether and how the taxonomic composition and metabolic activity of microbial community are affectedby timberline upshift.

The rapid development of high-throughput 16s rRNA gene sequencing has enabled fine-tuning assessment ofmicrobial taxonomic composition16. Meanwhile, development of a functional gene microarray (GeoChip) hasenabled quantitative, accurate and rapid detection of a large number of microbial genes17, making it possible to

OPEN

SUBJECT AREAS:

CLIMATE-CHANGEIMPACTS

MICROBIAL ECOLOGY

FOREST ECOLOGY

Received9 July 2014

Accepted29 December 2014

Published23 January 2015

Correspondence andrequests for materials

should be addressed toY.Y. (yangyf@

tsinghua.edu.cn) orY.Z. (zhangyg@caf.

ac.cn)

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 1

link the metabolic potentials of microbial communities with ecosys-tem processes18. Aiming to obtain insights in predicting changes inmicrobial communities and microbe-mediated soil biogeochemicalcycles for timberline upshift, we adopted 16s rRNA gene Illuminasequencing and GeoChip to analyze soil microbial communities ofthe coniferous forest and the shrubland, which are two vegetationbelts forming the timberline in the Shennongjia national naturalreserve (SNNR) located in Northwest Hubei province, China.SNNR is chosen because it is a sensitive region vulnerable to climatechanges19. It is one of the most biologically diverse areas in China anda member of UNESCO’s World Network of Biosphere Reserves,often known as ‘‘The Oriental Botanic Garden’’ and the natural bankof biological species.

Specifically, we are interested in the following scientific questions:(1) which environmental parameter best explained the formation ofmicrobial community? and (2) were there significant linkagesbetween microbial functional genes and soil nutrient storages, giventhe important role of microbes in driving biogeochemical cycles?

ResultsVegetation and soil geochemical parameters. The distance amongthe sampling plots within each vegetation zone ranged from 0.042–0.322 kilometers. Given the typical high soil heterogeneity, it wasunsurprising to detect considerable variations in the vegetationand soil geochemical parameters, which prompted us to test forgeneral differences between the shrubland and the coniferousforest (Table 1). We found that the number of plant species andplant diversity, calculated by Shannon index, were significantly (P, 0.001) higher in the coniferous forest (2.64) than in the shrubland(1.64). Fargesia murielae of the Poaceae family and Rhododendronoreodoxa of the Ericstarae family was the dominant species in theshrubland, while Abies fargesii of the Pinaceae family was thedominant species in the coniferous forest.

Many soil geochemical parameters, including soil organic carbon,labile organic carbon, dissolved organic carbon, total nitrogen, alkal-ine hydrolysis nitrogen, NO3

2-N, total potassium and rapidly avail-able phosphorus, were similar in both habitats. Soil temperature, amajor limiting factor for tree growth at the forest timberline20, wassignificantly higher (P 5 0.001) in the shrubland than the coniferousforest. This was consistent with the annual average air temperature,which was higher in the shrubland (4.62uC) than in the coniferousforest (4.00uC). For others, total phosphorus, total sulfur, NH4

1-Nand d15N were significantly (P , 0.010) higher in the shrubland,while moisture, pH, d13C, Al, and Fe were significantly (P , 0.010)higher in the coniferous forest. These results suggested that shrub-land soil was relatively richer in nutrients than the coniferous forest.

Soil microbial community of the coniferous forest and theshrubland. Microbial biomass, judged by soil microbial biomasscarbon and nitrogen, were ,40% higher at the significant level ofP , 0.002 in the shrubland than the coniferous forest (Table 1).At the taxonomic level, a total of 795,134 paired-end sequences,ranging from 14,928 to 50,849 reads per sample, were obtainedfrom 24 samples of these two sampling sites. Finally, random re-sampling was performed with 20,000 sequences per sample, result-ing in 55,440 OTUs at the 97% similarity level (SupplementaryFig. S1). The microbial taxonomic diversity, calculated by theSimpson index, was significantly (P 5 0.035) higher in theshrubland (Table 1). At the functional gene level, 45,357 geneswere detected in the shrubland samples, whereas 39,984 geneswere detected in the coniferous forest samples, resulting in asignificantly (P 5 0.001) higher functional diversity in theshrubland.

To further examine differences between these two sites, Principalcoordinates analysis was performed with the high-throughputsequencing and GeoChip data, respectively. Samples of the conifer-

Table 1 | Summary of environmental parameters analyzed by two-tailed t-test

Shrubland Coniferous forest P value

Plant parametersPlant species 12 (5) 47 (17) 0.001Plant diversity 1.64 (0.01) 2.64 (0.33) 0.001Microbial parametersTaxonomic diversity 248.49 (146.37) 121.58 (130.59) 0.035Functional diversity 126.19 (0.61) 124.18 (0.96) 0.001Soil microbial biomass carbon (mg/kg) 1457.39 (315.10) 1081.40 (247.75) 0.002Soil microbial biomass nitrogen (mg/kg) 169.20 (34.69) 120.34 (23.52) 0.001Soil biogeochemical parametersMoisture (%) 29.73 (2.73) 45.14 (4.23) 0.001Soil temperature (uC) 12.14 (0.35) 10.90 (0.50) 0.001Air annual average temperature (uC) 4.62 4.00 n/aSoil organic carbon (mg/kg) 59981.92 (12469.14) 63634.61 (16736.75) 0.793Labile organic carbon (mg/kg) 6093.65 (2622.35) 6441.53 (3422.21) 0.782Dissolved organic carbon (mg/kg) 283.27 (223.28) 206.90 (93.19) 0.282Total nitrogen (mg/kg) 4447.50 (1059.86) 4311.82 (852.72) 0.564Alkaline hydrolysis nitrogen (mg/kg) 404.51 (96.69) 350.57 (79.80) 0.092NH4

1-N (mg/kg) 19.29 (3.43) 15.26 (3.32) 0.006NO3

2-N (mg/kg) 11.02 (5.28) 18.42 (14.44) 0.147d13C (%) 225.98 (0.41) 225.44 (0.26) 0.001d15N (%) 4.61 (0.82) 3.00 (0.76) 0.001pH 4.43 (0.15) 5.07 (0.48) 0.001Rapidly available phosphorus (mg/kg) 8.39 (0.58) 11.41 (2.58) 0.580Total phosphorus (mg/kg) 1971.66 (1052.02) 604.50 (109.59) 0.001Total sulfur (mg/kg) 656.67 (259.32) 337.57 (72.84) 0.001Total potassium (mg/kg) 4281.21 (777.54) 4230.07 (1389.61) 0.739Al (mg/kg) 22572.10 (3378.57) 32555.23 (3378.57) 0.001Fe (mg/kg) 20766.62 (4067.16) 35420.78 (6583.12) 0.001Altitude (m) 2748.50 (15.89) 2541.36 (55.38) 0.001

P,0.050 is shown in bold. n/a: not applicable. Values in parenthesis represent standard deviation.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 2

ous forest and the shrubland were well separated from each other inboth datasets (Fig. 1), which was also verified by three non-paramet-ric multivariate statistical tests of dissimilarity (adonis, MRPP andanosim, P , 0.010) (data not shown).

The linkage between the microbial community and environment-al parameters. Pearson correlation analyses were carried out toidentify environmental parameters linking to soil microbial bio-mass. Soil temperature was positively correlated with soil microbialbiomass carbon (smbc) (r 5 0.54, P 5 0.006) and soil microbialbiomass nitrogen (smbn) (r 5 0.75, P 5 0.001) (Fig. 2A), implyingthat temperature might be a key factor in determining soil microbialbiomass. Soil NH4

1-N content was also positively correlated (P 5

0.002) with smbc and smbn (Supplementary Fig. S2A). But mostother environmental parameters, including soil pH and plantdiversity, were not significantly (P , 0.050) correlated.

We proceeded further to identify key environmental parameterslinking to microbial diversity. At the taxonomic level, soil temper-ature, but not NH4

1-N, was positively correlated (r 5 0.53, P 5

0.008) with microbial diversity (Fig. 2B). At the functional gene level,both soil temperature (Fig. 2B) and NH4

1-N (Supplementary Fig.S2A) were positively correlated (P , 0.050) with microbial func-tional diversity. Meanwhile, microbial taxonomic and functionaldiversity were positively correlated (P , 0.050) with soil microbialbiomass nitrogen (Supplementary Fig. S2B), showing a linkagebetween microbial biomass and diversity.

Canonical correspondence analysis (CCA) was performed toidentify major environmental parameters linking to soil microbialcommunity at the taxonomic or functional gene level. A total of 12environmental parameters (pH, moisture, soil temperature, NH4

1-N, total sulfur, total phosphorus, Al, Fe, altitude, plant diversity, d13Cand d15N) were selected as independent environmental parametersbased on variance inflation factor analysis, resulting in significant (P, 0.050) CCA models (Fig. 3). Soil temperature, moisture, pH andaltitude were the most important parameters shaping the microbialcommunity since they aligned well with the first axis, which repre-sented the major variations among microbial communities.

To single out the contribution of temperature to microbial com-munity variations, partial Mantel test was performed to correlate themicrobial community with temperature, soil parameters and plantcommunity. For taxonomic composition of microbial community,temperature was the most important control factor with the highestR-value (Table 2). For functional gene composition of microbial

community, temperature was the only factor that was significantlycorrelated.

Bacterial taxa groups. All of the detected OTUs were classified as 26bacterial phyla. Among them, Proteobacteria, Acidobacteria, Actino-bacteria and Verrucomicrobia were the most abundant phyla.Acidobacteria, Actinobacteria, Chloroflexi and Planctomycetes weresignificantly higher in the shrubland, while Proteobacteria wassignificantly higher in the coniferous forest (Supplementary Fig.S3). Proteobacteria, as the most abundant phylum, was composedof a-Proteobacteria, b-Proteobacteria, d-Proteobacteria and c-Proteobacteria. Among all of the detected Proteobacteria, b-Proteo-bacteria and a-Proteobacteria were the most abundant, which was ashigh as 72.62% of all Proteobacteria. b-Proteobacteria of theconiferous forest was significantly (P 5 0.001) and 192% higherthan that of the shrubland, while a-Proteobacteria was significantly(P 5 0.001) and 23% lower in the coniferous forest.

Selected gene categories of the shrubland and the coniferousforest. The response ratio analysis showed that genes associatedwith functional categories of organic remediation and soil benefitwere significantly more abundant in the coniferous forest at the95% confidence level (Supplementary Fig. S4). Genes associatedwith carbon cycle, sulfur cycle, bacterial phage, energy process andsoil borne pathogen were significantly less abundant, whereas genesassociated with nitrogen and phosphorus cycle remained similar.Detailed information of selected gene categories is described below.

Nitrogen (N) cycle. The total abundance of amoA genes associatedwith nitrification was higher in the coniferous forest (Fig. 4). Most ofthese amoA genes were derived from the archaea phylum ofCrenarchaeota, known to be highly abundant in temperate acidicforest soil21. By contrast, total abundances of nasA and nrfA genesassociated with N reduction and nirS/nirK genes associated withdenitrification were significantly lower, suggesting a shift of micro-bial functional potential from ammonium biosynthesis towardsnitrate biosynthesis. Consistently, the NO3

2-N content was 67.2%higher in the coniferous forest, albeit insignificantly, and the NH4

1-N content was significantly and 20.9% lower (Table 1).

We explored possible linkages between the NO32-N or NH4

1-Ncontent and N cycle genes associated with NO3

2-N or NH41-N

metabolism. The results showed that total abundances of nitrifica-tion genes (amoA and hao) were negatively correlated (r 5 20.46, P5 0.023) with NH4

1-N, and total abundances of denitrification genes

Figure 1 | Principal coordinates analysis (PCoA) of microbial community based on (A) high-throughput sequencing data and (B) GeoChip data. The

values for Axes 1, 2 and 3 are percentages of variation attributed to the corresponding axis.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 3

(nirS and nirK) were also negatively correlated (r 5 20.54, P 50.008) with NO3

2-N (Supplementary Fig. S5), demonstrating thatgene abundances associated with N cycle, representing microbialfunctional potentials, were closely linked to soil N contents.

Ammonium oxidization was controlled by a narrow group ofammonium-oxidizing archaea and bacteria22. Thus, we examinedammonium oxidizer OTUs from the sequencing data. Total abund-ance of the ammonium oxidizer was positively correlated to NO3

2-N(r 5 0.49, P 5 0.019) and negatively correlated to NH4

1-N (r 520.41, P 5 0.049) (Fig. 5A), supporting the important role of ammo-nium oxidizer in regulating soil N cycle. To verify it, we examinedbacterial amoA gene, which is the key functional gene of ammoniumoxidizers. Similar to the observations from OTU data, total abund-ance of bacterial amoA gene was also positively correlated to NO3

2-N (r 5 0.40, P 5 0.057) and negatively correlated to NH4

1-N (r 520.44, P 5 0.033) (Fig. 5B). Furthermore, total abundance of ammo-nium oxidizer was positively correlated with bacterial amoA geneabundance (r 5 0.38, P 5 0.064) (Fig. 5C).

Carbon (C) cycle. A large number of functional genes associated withfixation and degradation were detected. CODH (CO dehydrogenase)was significantly (P 5 0.050) more abundant in the coniferous forestbut rubisco (ribulose-1, 5-bisphosphate carboxylase/oxygenase) wasless abundant, suggesting that the CODH route was a preferred Cfixation route in the coniferous forest. Most of the abundant CODHgenes were derived from bacteria except a few (Sulfolobus acidocal-darius, Methanosarcina mazei and uncultured Crenarchaeote). Themost abundant CODH genes were derived from Bradyrhizobium sp.,Silicibacter sp., Aminobacter sp., Burkholderia sp., Ralstonia eutro-pha Solibacter usitatus, Mycobacterium smegmatis, Nocardioides sp.and uncultured bacteria, which were known to have a widedistribution23.

The total abundance of genes associated with C degradation wassignificantly lower in the coniferous forest samples (P 5 0.050).Total abundances of ara associated with hemicellulose degradation,

acetylglucosaminidase associated with chitin degradation and glxassociated with lignin degradation were higher in the coniferousforest, while the other genes were similar or lower (SupplementaryFig. S6A). The most abundant glx genes in the coniferous forest werederived from Phanerochaete chrysosporium, which has been welldocumented for its important role in lignin degradation24. Despitethese intrinsic differences of C fixation and degradation pathways,the net effect of C storage may be similar, as reflected in the obser-vation that the levels of soil organic carbon and dissolved organiccarbon were similar between the coniferous forest and the shrubland(Table 1).

Phosphorus (P) cycle. GeoChip 4.0 contained probes to target three Pcycle genes, ppk (polyphosphate kinase) for polyphosphate biosyn-thesis, phytase for phytate degradation and ppx (exopolyphospha-tase) for inorganic polyphosphate degradation. The total abundanceof ppx genes was significantly higher in the coniferous forest, whilethat of ppk genes was significantly lower (P , 0.050) (SupplementaryFig. S6B). No significant difference was found for phytase genes.Thus, microbial functional potentials are inclined towards polypho-sphate degradation, which was consistent with the higher level ofrapidly available phosphorus and the lower level of total phosphorusin the coniferous forest (Table 1).

Organic remediation. Abundant organic remediation genes in theconiferous forest included genes associated with biodegradation ofaromatics, chlorinated solvents, herbicides related compound, pes-ticides related compounds or other hydrocarbons. Among these,aromatic remediation genes were the most abundant. Consistently,more abundant aromatic components could be detected in conifer-rich sites25, which might induce the aromatics remediation genes inthe coniferous forest.

Correlation Networks of OTUs and functional genes. To explorepossible ecological interactions of microbial communities, we used a

Figure 2 | Correlations between (A) microbial biomass (indicated by soil microbial biomass carbon and soil microbial biomass nitrogen) and soiltemperature, and (B) microbial a-diversity (taxonomic and functional diversity) and soil temperature. White and black dots represent samples from the

shrubland and the coniferous forest, respectively.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 4

random matrix theory-based algorithm26 to reconstruct correlationnetworks from OTUs or selected GeoChip data. All of the resultingnetworks showed general features of ecological networks, such asscale free, small world and modular (Supplementary Table S1). Atboth taxonomic and functional gene levels, the coniferous forestnetworks had higher average connectivity and transitivity,indicative of tighter interactions/coupling within microbialcommunities. By contrast, the modularity, indicative of theresistance of systems to disturbance27, was lower in coniferousforest networks than that of shrubland. More positive interactions,which implicated cooperation28, were detected in the coniferousforest networks (98.19% in the OTU network and 92.44% in the

GeoChip network) than those in the shrubland networks (93.24%in the OTU network and 74.79% in the GeoChip network).

Only one hub of OTU networks was detected in the shrubland andthe coniferous forest, which was derived from the phylum ofAcidobacteria and Verrucomicrobia, respectively (SupplementaryFig. S7A). For GeoChip networks, 25 out of 34 hub nodes of theshrubland network were C cycle genes whereas 12 out of 20 hubnodes of the coniferous forest network were N cycle genes(Supplementary Fig. S7B), which was consistent with the finding thatrelative abundances of C and N cycle genes were higher and lower inthe shrubland samples, respectively (Supplementary Fig. S4).

To explore possible correlation between environmental factorsand microbial community, we included environmental factors inour correlation network based on random matrix theory. We foundthat temperature was a hub node, that is, temperature was one of thenodes with the highest connectivity with other nodes (Fig. 6). Someof the directly connected nodes to temperature were among the top10 nodes with the highest connectivity, suggesting that temperatureand its directly connected nodes play an important role in maintain-ing the network structure and topology.

DiscussionInvestigating soil microbial communities is important for under-standing microbe-mediated biogeochemical cycle and ecosystemfunctioning29. A main goal of this study is to identify the envir-

Figure 3 | Canonical correspondence analysis (CCA) of (A) high-throughput sequencing data and (B) GeoChip data with environmentalparameters. TS, total sulfur; TP, total phosphorus.

Table 2 | The relationships of microbial community structure to environmental parameters revealed by partial Mantel tests

In association with: controlling for:microbial communities

Temperature a Soil b Plant c

Soil Plant Temperature Plant Temperature Soil

rM(P) rM(P) rM(P) rM(P) rM(P) rM(P)

OTUs 0.54 (0.001) 0.60 (0.001) 0.41 (0.001) 0.46 (0.001) 0.22 (0.056) 0.14 (0.094)Functional genes 0.13 (0.031) 0.16 (0.014) 0.11 (0.215) 0.14 (0.160) 0.05 (0.375) 0.03 (0.417)aTemperature included soil temperature and air annual temperature.bSoil parameters included soil organic carbon, labile organic carbon, dissolved organic carbon, total nitrogen, alkaline hydrolysis nitrogen, NH4

1-N, NO32-N, d13C, d15N, soil pH, Rapidly available

phosphorus, total phosphorus, total sulfur, total potassium, Al, Fe.cPlant parameters was the plant community structure, composed of the significant value of each specie in every sampling site.

Figure 4 | The fold change of N cycle genes between the shrubland andthe coniferous forest. The percentage of each gene indicated the relative

fold changes, with red and green colors indicating the higher and lower

signal intensity of each detected gene in the coniferous forest, respectively.

This percentage was calculated by the sum of the signal intensity of each

detected gene divided by the total sum of signal intensity of all detected N

cycle genes, and weighted by the fold change of each gene for the coniferous

forest over the shrubland. ***P , 0.010; **P , 0.050.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 5

onmental factor(s) that best explain microbial community variation.It has been proposed that above-ground plant communities drivebelow-ground microbial diversity30. Previous studies have indicatedthat plants affect microbial communities via rhizodeposits and exu-dates and provide organic C and N to the soil microbial commun-ity31,32. The different vegetation types could select for distinct soilmicroorganisms due to formation of a variety of microhabitats thatsupport a diverse collection of species33,34. Thus, we examinedwhether soil microbial functional diversity was affected by plantdiversity. However, significantly higher microbial taxonomic andfunctional diversity were detected in the shrubland than the conifer-ous forest (Table 1). Considering higher plant diversity, species num-ber and biomass in the coniferous forest, this finding suggested thatplant diversity might not be the determinant of microbial diversity,which was consistent with several previous studies showing that soilmicrobial community was uncorrelated with plant diversity35–37. Theinconsistency between plant diversity and microbes could beascribed to different ecophysiological traits of plant species, whichwould exert strong effects on soil biological properties38,39. Thus, theplant community composition is more likely to affect soil microbialcommunity composition than plant diversity35,40,41. In addition, thiscould also be explained by a chance effect that a keystone speciespresent in the plant community results in greater effects on soilprocessing than the contribution of the total plant diversity42.Notably, the dominant species were different between the shrublandand the coniferous forest. The plant litter of Abies fargesii in theconiferous forest with relatively higher C/N ratios provides poor

nutrients for microbial growth43, and this may have a tendency toreduce microbial diversity and activity. On the contrary, Fargesiamurielae, one of the dominate plants in the shrubland, would buffertemperature extremes and modify the local microenvironments andsoil quality44, which might be helpful for microbial survival.

Temperature has been shown to be the major factor in timberlineformation9. Here we showed that soil temperature, among all ofmeasured environmental parameters, showed the most significantand extensive linkages with microbial biomass, microbial diversityand composition at both taxonomic and functional gene levels.Correlation networking analysis also showed that temperature wasan influential environmental parameter. Therefore, temperature wasthe best predictor for microbial community variations at the timber-line. Our study showed that the temperature was significantly lowerin coniferous forest, which was consistent with previous studiesshowing that the timberline forest typically had a colder temperaturethan its adjacent vegetation at higher altitudes2,45,46, since the closedforest canopy protects the soil from sunrays and results in a lowertemperature. Temperature can affect metabolic rates via the kineticsof the biological process, famously coined as ‘‘the Red Queen runsfaster when she is hot’’, and consequently contributes to organismdifferentiation and diversification47,48. However, it is also possiblethat temperature indirectly affects microbial community by control-ling soil nutrient availability, which further affects soil microbialcommunity49. A comparison among the shrubland, grassland, farm-land and reforested land in an adjacent region to our study sitesshowed that the shrubland had the highest soil quality value50.

Figure 5 | Correlations between (A) NH41-N, NO3

2-N and total abundance of ammonium oxidizer, (B) NH41-N, NO3

2-N and total abundance ofbacterial amoA gene, and (C) abundance of ammonium oxidizer and bacterial amoA gene. White and black dots represent samples from the shrubland

and the coniferous forest, respectively.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 6

Similar studies on alpine ecosystems showed that treeless soil athigher altitudes had more labile C, greater microbial activity thansoil of closed forest canopy51,52. Consistently, higher soil nutrientcontents of dissolved organic carbon, available nitrogen, total phos-phorus and total sulfur in the shrubland provided a suitable envir-onment for microbial communities, as indicated by higher microbialbiomass and diversity. CCA showed that microbial community com-positions were correlated with d15N, total phosphorus, total sulfur; Aland Fe (Fig. 3), in addition to correlations between microbial bio-mass, functional diversity and NH4

1-N (Supplementary Fig. S2).Among them, N appeared to be a key factor affecting microbialcommunity composition, diversity and productivity in many N-lim-ited terrestrial ecosystems53. The correlation between microbial bio-mass and NH4

1-N was consistent with precious study showing thatN limitation was an influential factor in affecting microbial com-munity54. Meanwhile, our study showed that low microbial divers-ities were detected in sampling sites with high C/N ratios, which wasconsistent with the finding that microbial community structure andactivity were negatively coupled to the soil C/N ratio43. This indicatedthat an upward expansion of forest resulting from global warmingwould lead to changes in soil organic quality and the activities of theunderground communities.

A parallel 16s rRNA and shotgun sequencing study on tallgrassprairies showed a strong positive correlation between taxonomic andfunctional gene diversity55, suggesting a low degree of functionalredundancy. By contrast, we did not detect strong correlationbetween overall taxonomic and functional gene diversity, whichmight be attributed to the differences in techniques or ecosystems.However, there was a positive correlation between bacterial amoAgene abundance and ammonium oxidizers (Fig. 5C), a functionallynarrow group utilizing the conversion of ammonia into nitrite as the

sole energy source22. It can thus be predicted that reduction inammonium oxidizer species is associated with decreases in the func-tional potential of bacterial ammonium oxidization contained withinthese soil communities.

Correlation networking analyses show important details of com-munity assembly rules reflecting ecological processes such as coop-eration, competition, habitat filtering and historical effects, and canrepresent mathematical interaction/coupling among different popu-lations and/or functional genes that regulate system functions56. Theshorter path length and higher average connectivity and transitivityof coniferous forest networks suggested that microbial interaction/coupling was high. The high interaction might be ascribed to certaindeterministic processes such as habitat filtering, reducing the spreadof trait values and reflecting shared ecological tolerances57, which wasconsistent with the observations of more positive interactions in theconiferous forest networks. Modularity helps control disturbances bycompartmentalizing social-ecological systems58, the higher modular-ity of the shrubland network indicated that the microbial systemwould be more resistant to changes, both at the taxonomic andfunctional level, which might be ascribed to higher microbial divers-ity of shrubland (Supplementary Table S1).

Profiling microbial communities in the habitats bordering thetimberline is crucial for predicting the dynamics of microbial com-munity changes and ecosystem functioning, since the trend of tim-berline upshift resulting from climate warming is likely to continue.Our results showed that temperature was the best predictor formicrobial community formation and there were clear linkagesbetween microbial functional potentials and soil biogeochemistrycycles of the timberline. Based on these results, we predict that tim-berline upshift resulting from global warming would cause distinctchanges in microbial communities and soil C and N pools. However,

Figure 6 | Network interactions between environmental parameters and microbial taxonomic community. The red node represents temperature. The

green ones represent the nodes directly connected to temperature, bigger diameter of which indicates higher connectivity. The purple ones were indirectly

connected OTUs.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 7

the general trend of timberline shift is modified by local, regional andtemporal variations59. It would be interesting to expand the observa-tions in this study to other timberlines or conduct time-series experi-ments in order to test the generality of the observations.

MethodsSite and sampling. The study site is located at Shennongjia National Nature Reserve(SNNR, 110u14’–110u18’ E, 31u26’–31u28’ N) of Shennongjia Mountain, NorthwestHubei province, Central China. This area is heavily influenced by subtropicalmonsoon, characterized by wet (monthly precipitation . 250 mm) and hot weatherin summer (monthly mean temperature . 25uC) but dry (monthly precipitation ,

10 mm) and cold (monthly mean temperature , 24uC) weather in winter19. Theannual mean air temperature and precipitation of this area are 7.2uC and 1500 mm,respectively. The vegetation belts forming the timberline were the shrubland and theconiferous forest. As many other landscape lines, the timberline is a well-definedboundary from a distance but a gradual transition with patchy patterns upon closeinspection. To manifest differences between the shrubland and the coniferous forest,we chose typical shrubland between the altitudes of 2720 and 2776 m and theconiferous forest between the altitudes of 2456 and 2632 m to collect soil samples.

Soil samples of the shrubland and the coniferous forest were collected in August,2011. At each site, twelve plots (20 m 3 20 m) comprising typical vegetation and soilattributes were selected for sampling. At each plot, ten to fifteen soil cores at a depth of0–10 cm were taken, mixed thoroughly and sieved with 2 mm mesh to remove rootsand stones. Soil samples were kept on ice when transporting to laboratory, dividedinto two subsamples and stored at either 4uC for soil biogeochemical measurementsor 280uC for DNA extraction.

Soil and vegetation parameter measurements. The species, number and canopy ofeach tree or shrub, the diameter of breast height and height of trees were surveyed ateach plot. The air annual average temperature (from 1950 to 2000) was calculatedwith long-term records of daily air temperature by WorldClim (Version 1.4)60. Soiltemperature at each plot was measured by a Hobo Temperature instrument at thedepths of 0–15 cm, respectively. The other soil geochemical parameters (soilmoisture, soil pH, soil organic carbon, total nitrogen, available phosphorous, availablepotassium, water, dissolved organic carbon, labile organic carbon contents, d13C, d15Nand soil enzyme activities of cellulose, glucanase, polyphenol oxidase, sucrase andamylase) were measured as previously described61. A rapid chloroform-fumigationextraction method was used to measure soil microbial biomass carbon and nitrogenas previously described62. Shannon-Weaver index (H9) was used to evaluate the plantdiversity.

DNA extraction, purification, quantification. Soil DNA was extracted by freeze-grinding mechanical lysis as previously described63. Freshly extracted DNA waspurified twice using 0.5% low melting point agarose gel followed by phenol-chloroform-butanol extraction. DNA quality and quantity were assessed by the ratiosof 260 nm/280 nm and 260 nm/230 nm, and final DNA contents were quantifiedwith a PicoGreen method using a FLUO star Optima (BMG Labtech, Jena, Germany).

Illumina sequencing and data processing. 16s rDNA genes was amplified using acommon primer pair targeting v4 region (Forward primer, 515F, 59-GTGCCAGCMGCCGCGGTAA-39; reverse primer, 806R, 59-GGACTACHVGGGTWTCTAAT-39) combined with adapter sequences andbarcode sequences64,65. PCR amplification was performed in a 25 ml reaction,containing 2.5 ml 10 3 AccuPrime PCR buffer II (Invitrogen, Grand Island, NY), 1 mlof each primer (10 mM), 5 ml template DNA (2 ng/ml) and 0.1 ml AccuPrime HighFidelity Taq Polymerase. The reaction mixtures were subjected to a denaturation at94uC for 1 min, followed by 30 cycles of 94uC for 20 s, 53uC for 25 s and 68uC for45 s, and a final extension at 68uC for 10 min. Reactions were performed in triplicatesfor each sample to minimize potential biases from amplification66. The triplicateproducts from the first step PCR were pooled together and purified throughQIAquick Gel Extraction Kit (Qiagen, Valencia, CA), eluted in 50 ml water, andaliquoted into three PCR tubes. A second round PCR was then performed in a 25 mlreaction [2.5 ml 10 3 AccuPrime PCR buffer II, 1 ml of each primer (10 mM), 0.1 mlAccuPrime High Fidelity Taq Polymerase, and 15 ml aliquot of the first-roundpurified PCR product] with triplicates for each sample, using phasing primers withIllumina adapters, target primers, spacers and barcodes on reverse primers. Twentycycles of the amplifications were performed following the same program for the firstround PCR. PCR products from triplicate reactions were pooled together andquantified with PicoGreen. Sample denaturation was carried out by mixing 10 ml ofcombined PCR products and 10 ml 0.2 N fresh NaOH. The chilled Illumina HT1buffer was added to the denatured DNA to make a 20 pM library, which was furtherdiluted to 15 pM by adding HT1 buffer and mixed with a PhiX DNA library. Finally,the 600 ml mixture was loaded into the MiSeq reagent cartridge and run on Miseq(Illumina, San Diego, CA) for 2 3 250 bp paired-end sequencing.

Raw sequences were separated to samples using barcodes and with permission ofup to one mismatch. Quality trimming was done using Btrim67. Forward and reversereads were merged into full length sequences by FLASH68. Sequences were removed ifthey were too short or contained ambiguous bases. Random re-sampling was per-formed with 20,000 sequences per sample. The operational taxonomic units (OTUs)were classified using UCLUST at the 97% similarity level, and singletons were

removed. The taxonomic assignment was conducted by RDP classifier69 with minimal50% confidence estimates.

GeoChip hybridization and data processing. Purified DNA was labeled with Cy3dye, dried, rehydrated and hybridized with GeoChip 4.0 as previously described18.After purification, GeoChip microarrays were scanned with a 100% laser power and100% photomultiplier tube with a NimbleGen MS 200 Microarray Scanner (Roche,Basel, Switzerland), and poor-quality spots with a signal-to-noise ratio of less than 2.0were discarded prior to statistical analyses.

Signal intensities were quantified and processed using the analysis pipeline aspreviously described70. Then processed GeoChip data were analyzed using the fol-lowing steps: (i) removing genes detected in no more than 5 out of 12 samples fromthe same vegetation type; (ii) normalizing the signal intensity of each spot by dividingthe signal intensity by the total intensity of the microarray followed by multiplying bya constant; and (iii) transforming data to the natural logarithmic form.

The Simpson’s reciprocal index (1/D) was used to evaluate microbial taxonomicand functional diversity. Principal coordinates analysis (PCoA) was used to deter-mine changes of overall microbial community structure, both the high-throughputsequencing data and GeoChip data. Bray-Curtis distances were used to calculate thedissimilarity matrices from the high-throughput sequencing data and GeoChip data.Canonical correspondence analysis was used to explore linkages between microbialcommunity structure and environmental parameters. The variance inflation factor(VIF) was used for step-wise removal of redundant parameters in CCA modeling17.Partial Mantel test was performed to correlate the microbial community (for high-throughput sequencing data, OTUs detected in more than 10 out of 24 samples wasused) with the temperature, soil parameters and plant community. All of the analyseswere performed by functions in the Vegan package (v.1.15–1) in R (v.3.0.1).

Network construction via the RMT-based method. Using the random matrix theory(RMT)-based method to automatically define a threshold, Pearson correlationnetworks were constructed from OTUs or GeoChip data as previously described.OTUs or selected functional genes (carbon and nitrogen cycle genes) detected in lessthan nine out of twelve replicates from the coniferous forest and the shrubland wereremoved to ensure reliable correlations. To explore the ecological interactionsbetween environmental parameters and microbial community, OTUs andenvironmental parameters detected in less than eighteen out of twenty-four sampleswere removed to ensure reliable correlations. To characterize the modularityproperty, each network was separated into modules by the fast greedy modularityoptimization. Various indices, including the average degree, average clusteringcoefficient, average path distance, geodesic efficiency, harmonic geodesic distance,density and transitivity, and connectedness, were used to describe overall topologicalproperties of different networks. The topological roles of different nodes were dividedinto the following four subcategories by within-module connectivity (Zi) and among-module connectivity (Pi): (i) peripheral nodes are defined as nodes with Zi # 2.5 andPi # 0.62; (ii) connectors are defined as nodes with Zi # 2.5 and Pi . 0.62; (iii)module hubs are defined as nodes with Zi . 2.5 and Pi # 0.62; and (iv) network hubsare defined as nodes with Zi . 2.5 and Pi . 0.62.

1. Gruber, A., Pirkebner, D., Oberhuber, W. & Wieser, G. Spatial and seasonalvariations in mobile carbohydrates in Pinus cembra in the timberline ecotone ofthe Central Austrian Alps. Eur J Forest Res 130, 173–179 (2011).

2. Korner, C. & Paulsen, J. A world-wide study of high altitude treeline temperatures.J. Biogeogr. 31, 713–732 (2004).

3. Malanson, G. P. et al. Mountain treelines: a roadmap for research orientation.Arct. Antarct. Alp. Res. 43, 167–177 (2011).

4. Theurillat, J.-P. & Guisan, A. Potential impact of climate change on vegetation inthe European Alps: a review. Clim. Change 50, 77–109 (2001).

5. Panigrahy, S., Anitha, D., Kimothi, M. & Singh, S. Timberline change detectionusing topographic map and satellite imagery. Trop. Ecol. 51, 87–91 (2010).

6. Graumlich, L. J., Waggoner, L. A. & Bunn, A. G. in Global change and mountainregions 501–508 (Springer, 2005).

7. Holtmeier, F. K. Mountain timberlines. Vol. 36 (Springer, 2009).8. Holtmeier, F. K. & Broll, G. Sensitivity and response of northern hemisphere

altitudinal and polar treelines to environmental change at landscape and localscales. Global Ecol. Biogeogr. 14, 395–410 (2005).

9. Harsch, M. A., Hulme, P. E., McGlone, M. S. & Duncan, R. P. Are treelinesadvancing? A global meta-analysis of treeline response to climate warming. Ecol.Lett. 12, 1040–1049 (2009).

10. Lingua, E., Cherubini, P., Motta, R. & Nola, P. Spatial structure along an altitudinalgradient in the Italian central Alps suggests competition and facilitation amongconiferous species. J Vet Sci 19, 425–436 (2009).

11. Motta, R. & Nola, P. Growth trends and dynamics in sub-alpine forest stands inthe Varaita Valley (Piedmont, Italy) and their relationships with human activitiesand global change. J Vet Sci 12, 219–230 (2001).

12. Kammer, A. et al. Treeline shifts in the Ural mountains affect soil organic matterdynamics. Global Change Biol. 15, 1570–1583 (2009).

13. Wieser, G. & Stohr, D. Net ecosystem carbon dioxide exchange dynamics in aPinus cembra forest at the upper timberline in the central Austrian Alps. Phyton45, 233–242 (2005).

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 8

14. Withington, C. L. & Sanford Jr, R. L. Decomposition rates of buried substratesincrease with altitude in the forest-alpine tundra ecotone. Soil Biol. Biochem. 39,68–75 (2007).

15. Reeve, J. R. et al. Effects of soil type and farm management on soil ecologicalfunctional genes and microbial activities. ISME J 4, 1099–1107 (2010).

16. Shendure, J. & Ji, H. Next-generation DNA sequencing. Nat. Biotechnol. 26,1135–1145 (2008).

17. He, Z. et al. GeoChip: a comprehensive microarray for investigatingbiogeochemical, ecological and environmental processes. ISME J 1, 67–77 (2007).

18. Yang, Y. et al. Responses of the functional structure of soil microbial communityto livestock grazing in the Tibetan alpine grassland. Global Change Biol. 19,637–648 (2013).

19. Dang, H., Zhang, K., Zhang, Y., Tong, X. & Zhang, Q. Regeneration dynamics ofsubalpine fir (Abies fargesii) forest across the altitudinal range in the ShennongjiaMountains, central China. J Plant Ecol 6, 36–47 (2013).

20. Alvarez-Uria, P. & Korner, C. Low temperature limits of root growth in deciduousand evergreen temperate tree species. Funct. Ecol. 21, 211–218 (2007).

21. Kemnitz, D., Kolb, S. & Conrad, R. High abundance of Crenarchaeota in atemperate acidic forest soil. FEMS Microbiol Ecol 60, 442–448 (2007).

22. Kowalchuk, G. A. & Stephen, J. R. Ammonia-oxidizing bacteria: a model formolecular microbial ecology. Annu Rev Microbiol 55, 485–529 (2001).

23. King, G. M. & Weber, C. F. Distribution, diversity and ecology of aerobic CO-oxidizing bacteria. Nat Rev Microbiol 5, 107–118 (2007).

24. Kersten, P. J. & Cullen, D. Cloning and characterization of cDNA encodingglyoxal oxidase, a H2O2-producing enzyme from the lignin-degradingbasidiomycete Phanerochaete chrysosporium. P Natl Acad Sci USA 90,7411–7413 (1993).

25. Ussiri, D. A. & Johnson, C. E. Characterization of organic matter in a northernhardwood forest soil by 13C NMR spectroscopy and chemical methods. Geoderma111, 123–149 (2003).

26. Zhou, J., Deng, Y., Luo, F., He, Z. & Yang, Y. Phylogenetic molecular ecologicalnetwork of soil microbial communities in response to elevated CO2. MBio 2,e00122–11 (2011).

27. Scheffer, M. et al. Anticipating critical transitions. Science 338, 344–348 (2012).28. Zhang, Y., Zhao, Z., Dai, M., Jiao, N. & Herndl, G. J. Drivers shaping the diversity

and biogeography of total and active bacterial communities in the South ChinaSea. Mol. Ecol. 23, 2260–2274 (2014).

29. Falkowski, P. G., Fenchel, T. & Delong, E. F. The microbial engines that driveEarth’s biogeochemical cycles. Science 320, 1034–1039 (2008).

30. Mitchell, R. J. et al. Is vegetation composition or soil chemistry the best predictorof the soil microbial community? Plant Soil 333, 417–430 (2010).

31. Wardle, D. A. Communities and Ecosystems: Linking the Aboveground andBelowground Components. (Princeton Univ. Press, 2002).

32. Eisenhauer, N. et al. Plant diversity effects on soil microorganisms support thesingular hypothesis. Ecology 91, 485–496 (2010).

33. Faoro, H. et al. Influence of soil characteristics on the diversity of bacteria in theSouthern Brazilian Atlantic Forest. Appl. Environ. Microbiol. 76, 4744–4749(2010).

34. Zhou, J., Xia, B., Huang, H., Palumbo, A. V. & Tiedje, J. M. Microbial diversity andheterogeneity in sandy subsurface soils. Appl. Environ. Microbiol. 70, 1723–1734(2004).

35. Broughton, L. & Gross, K. Patterns of diversity in plant and soil microbialcommunities along a productivity gradient in a Michigan old-field. Oecologia 125,420–427 (2000).

36. Kowalchuk, G. A., Buma, D. S., de Boer, W., Klinkhamer, P. G. L. & van Veen, J. A.Effects of above-ground plant species composition and diversity on the diversityof soil-borne microorganisms. Anton Van Leeuw 81, 509–520 (2002).

37. Zak, D. R., Holmes, W. E., White, D. C., Peacock, A. D. & Tilman, D. Plantdiversity, soil microbial communities, and ecosystem function: are there anylinks? Ecology 84, 2042–2050 (2003).

38. Singh, B. K., Dawson, L. A., Macdonald, C. A. & Buckland, S. M. Impact of bioticand abiotic interaction on soil microbial communities and functions: A fieldstudy. Appl Soil Ecol 41, 239–248 (2009).

39. Vries, F. T. et al. Abiotic drivers and plant traits explain landscape-scale patternsin soil microbial communities. Ecol. Lett. 15, 1230–1239 (2012).

40. Yergeau, E., Lawrence, J. R., Waiser, M. J., Korber, D. R. & Greer, C. W.Metatranscriptomic analysis of the response of river biofilms to pharmaceuticalproducts, using anonymous DNA microarrays. Appl. Environ. Microbiol. 76,5432–5439 (2010).

41. Zinger, L. et al. Global patterns of bacterial beta-diversity in seafloor and seawaterecosystems. PLoS ONE 6, e24570 (2011).

42. Ladygina, N. & Hedlund, K. Plant species influence microbial diversity and carbonallocation in the rhizosphere. Soil Biol. Biochem. 42, 162–168 (2010).

43. Djukic, I., Zehetner, F., Mentler, A. & Gerzabek, M. H. Microbial communitycomposition and activity in different Alpine vegetation zones. Soil Biol. Biochem.42, 155–161 (2010).

44. He, Y., Xu, X., Kueffer, C., Zhang, X. & Shi, P. Leaf litter of a dominant cushionplant shifts nitrogen mineralization to immobilization at high but not lowtemperature in an alpine meadow. Plant Soil 383, 415–426 (2014).

45. Korner, C., Paulsen, J. & Pelaez-Riedl, S. A bioclimatic characterisation of Europe’salpine areas. (Springer, 2003).

46. Hadley, J. & Smith, W. Influence of krummholz mat microclimate on needlephysiology and survival. Oecologia 73, 82–90 (1987).

47. Davidson, E. A. & Janssens, I. A. Temperature sensitivity of soil carbondecomposition and feedbacks to climate change. Nature 440, 165–173 (2006).

48. Allen, A. P., Brown, J. H. & Gillooly, J. F. Global biodiversity, biochemical kinetics,and the energetic-equivalence rule. Science 297, 1545–1548 (2002).

49. Mitchell, R. J. et al. Explaining the variation in the soil microbial community: dovegetation composition and soil chemistry explain the same or different parts ofthe microbial variation? Plant Soil 351, 355–362 (2012).

50. Fu, B. J., Liu, S. L., Chen, L. D., Lv, Y. H. & Qiu, J. Soil quality regime in relation toland cover and slope position across a highly modified slope landscape. Ecol. Res.19, 111–118 (2004).

51. Sjogersten, S., Turner, B. L., Mahieu, N., Condron, L. M. & Wookey, P. A. Soilorganic matter biochemistry and potential susceptibility to climatic change acrossthe forest-tundra ecotone in the Fennoscandian mountains. Global Change Biol.9, 759–772 (2003).

52. Neff, J. C. & Hooper, D. U. Vegetation and climate controls on potential CO2,DOC and DON production in northern latitude soils. Global Change Biol. 8,872–884 (2002).

53. Zancarini, A. et al. Soil nitrogen availability and plant genotype modify thenutrition strategies of M. truncatula and the associated rhizosphere microbialcommunities. PLoS ONE 7, e47096 (2012).

54. Thebault, A. et al. Nitrogen limitation and microbial diversity at the treeline. Oikos123, 729–740 (2014).

55. Fierer, N. et al. Reconstructing the Microbial Diversity and Function of Pre-Agricultural Tallgrass Prairie Soils in the United States. Science 342, 621–624(2013).

56. Fuhrman, J. A. Microbial community structure and its functional implications.Nature 459, 193–199 (2009).

57. Cornwell, W. K., Schwilk, D. W. & Ackerly, D. D. A trait-based test for habitatfiltering: convex hull volume. Ecology 87, 1465–1471 (2006).

58. Carpenter, S. R. et al. General resilience to cope with extreme events. Sustainability4, 3248–3259 (2012).

59. Holtmeier, F.-K. & Broll, G. Treeline advance-driving processes and adversefactors. Landscape Online 1, 1–33 (2007).

60. Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very highresolution interpolated climate surfaces for global land areas. Int J Climatol 25,1965–1978 (2005).

61. Bao, S. D. Soil and agricultural chemistry analysis. Agriculture Publication, Beijing,355–356 (2000).

62. Witt, C., Gaunt, J. L., Galicia, C. C., Ottow, J. C. & Neue, H.-U. A rapidchloroform-fumigation extraction method for measuring soil microbial biomasscarbon and nitrogen in flooded rice soils. Biol. Fertility Soils 30, 510–519 (2000).

63. Zhou, J., Bruns, M. A. & Tiedje, J. M. DNA recovery from soils of diversecomposition. Appl. Environ. Microbiol. 62, 316–322 (1996).

64. Caporaso, J. G. et al. Global patterns of 16s rRNA diversity at a depth of millions ofsequences per sample. P Natl Acad Sci USA 108, 4516–4522 (2011).

65. Caporaso, J. G. et al. Ultra-high-throughput microbial community analysis on theIllumina HiSeq and MiSeq platforms. ISME J 6, 1621–1624 (2012).

66. Zhou, J. et al. Reproducibility and quantitation of amplicon sequencing-baseddetection. ISME J 5, 1303–1313 (2011).

67. Kong, Y. Btrim: a fast, lightweight adapter and quality trimming program for next-generation sequencing technologies. Genomics 98, 152–153 (2011).

68. Magoc, T. & Salzberg, S. L. FLASH: fast length adjustment of short reads toimprove genome assemblies. Bioinformatics 27, 2957–2963 (2011).

69. Wang, Q., Garrity, G. M., Tiedje, J. M. & Cole, J. R. Naive Bayesian classifier forrapid assignment of rRNA sequences into the new bacterial taxonomy. Appl.Environ. Microbiol. 73, 5261–5267 (2007).

70. He, Z. et al. GeoChip 3.0 as a high-throughput tool for analyzing microbialcommunity composition, structure and functional activity. ISME J 4, 1167–1179(2010).

AcknowledgmentsThis research was supported by grants to Yuguang Zhang from the public welfare project ofthe national scientific research institution (CAFRIFEEP201101), to Yunfeng Yang fromMajor Science and Technology Program for Water Pollution Control and Treatment(2013ZX07315-001-03), the Strategic Priority Research Program of the Chinese Academyof Sciences (XDB15010102), the National Key Basic Research Program of China(2013CB956601), National High Technology Research and Development Program ofChina (2012AA061401) and National Science Foundation of China (41471202 &41171201), to Ye Deng from State Key Laboratory of Forest and Soil Ecology (Grant No.LFSE2014-02) and the Strategic Priority Research Program of the Chinese Academy ofSciences (Grant XDB15010302), and to Jizhong Zhou from the US National ScienceFoundation (EF-1065844).

Author contributionsY.Y., Y.Z., J.Z. and D.L. designed experiments. J.D. and Y.Y. wrote the main manuscripttext. J.D., Y.D. and X.S. analyzed the data. J.C., H.L., C.Y., T.Y. and J.V.N. performed theexperiments. All authors reviewed the manuscript.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 9

Additional informationData Accession Microarray data have been deposited to GEO databases (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc5GSE58433). Sequencing data are currently beingdeposited to GEO databases.

Supplementary information accompanies this paper at http://www.nature.com/scientificreports

Competing financial interests: The authors declare no competing financial interests.

How to cite this article: Ding, J. et al. Integrated metagenomics and network analysis of soilmicrobial community of the forest timberline. Sci. Rep. 5, 7994; DOI:10.1038/srep07994(2015).

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License. The images or other third party material inthis article are included in the article’s Creative Commons license, unless indicatedotherwise in the credit line; if the material is not included under the CreativeCommons license, users will need to obtain permission from the license holderin order to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/4.0/

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 5 : 7994 | DOI: 10.1038/srep07994 10